The present invention relates to an adaptive digital filter which approximates with a nonrecursive digital filter the characteristics of a transmission system which is unknown or changes over time.
In a conventional adaptive digital filter, a nonrecursive digital filter with variable coefficient parameters is connected in parallel with a transmission system whose characteristics are unknown, and the unknown characteristics are approximated or simulated by the nonrecursive digital filter. In such an adaptive digital filter, a sequential correction algorithm is used to add a correction value to a coefficient parameter at a given time to obtain a coefficient parameter at a subsequent time. For example, if a speech signal is received at an input terminal of a transmission system including a loudspeaker and a microphone and transmitted from an output terminal thereof, the nonrecursive digital filter uses the adaptive algorithm to approximate feedback characteristics of the speech signal fed back to the microphone from the loudspeaker.
When the characteristics of the transmission system, i.e., the feedback characteristics between the loudspeaker and the microphone are abruptly changed, a certain period of time is required for the nonrecursive digital filter to approximate the changed characteristics of the transmission system with the adaptive algorithm. This approximation period lasts until an error signal eK, representing the difference between both characteristics, appears at the output terminal. For example, when a circuit such as an echo canceller between the input and output terminals serves as part of a loop and the level of the error signal eK is high, howling occurs. The nonrecursive digital filter cannot completely approximate the unknown characteristics of the transmission system, which change as a function of time, and thus an error is always present. This error must be minimized.